Anda di halaman 1dari 6

Proceedings of International Joint Conference on Neural Networks, Orlando, Florida, USA, August 12-17, 2007

WaterBalloons: A Hybrid Watershed Balloon Snake Segmentation


Issam Dagher and Kamal El Tom

basins, and the final union of all the lines will define the
Abstract- In this paper, a new image segmentation watershed segmentation.
technique called WaterBalloons is introduced. It If we imagine a hole in each local minimum and
combines both Watershed segmentation and the Active immerse the topographic surface in water, the water starts
Contour Model known as Balloon Snake. The filling all the catchment basins progressively. There will be
Watershed transform has a major problem of over- an instant at which water coming from two different
segmentation. Solutions like Region merging, use of minima would meet and merge the two basins. To avoid
markers, use of multi-scales have been proposed. These that different basins merge, a dam is built at each point of
approaches led to other problems such as under- contact. Following the flooding or immersion process, the
segmentation. The Balloon Snake in an innovative union of all those dams constitutes the watersheds. The
approach that detects salient objects in an image. But watershed transform denotes the assignment of all points
in general Snakes are very sensitive to initialization and in a distinct catchment basin with the same label, while a
need user interactions and a-priori knowledge of special label is assigned to all point of the watershed.
objects to segment. WaterBalloons provide the Efficient algorithms have been proposed to simulate the
advantage of reducing watershed over-segmentation immersion process [3], [4].
problems while preventing under-segmentation and
ensure automatic initialization of traditional snakes. In In [5] the notions of basins and contours dynamics are
addition, a method for parameter optimization of the introduced, which require interactive dynamics
proposed hybrid snake is introduced based on energy thresholding. Multiscale analysis [6] of intensity minima
transitions tracking. in the gradient magnitude image provides a mechanism for
imposing a scale-based hierarchy on the watersheds
regions, which can be used to label watershed boundaries
according to their scale. The disadvantage of this method
i. INTRODUCTION is that severe Gaussian blurring through scale space results
in dislocated or perturbed edges.
Image segmentation is the process of distinguishing The advantage of the watershed transform is that it
objects of interest from a given image. In this section, the produces closed and adjacent contours including all image
following segmentation techniques are introduced: edges. Often the watershed produces a severe over-
The watershed transform, the Active contour models segmentation, but the crucial point is that all important
(Snakes), and the Balloon Snakes. object boundaries are included, and the task is reduced to
eliminating the undesired ones.
A. The watershed transform Some solutions of the over-segmentation are:
The watershed transform is the basic segmentation tool in
mathematical morphology [1], [2]. The watershed concept 1) Region Merging:
comes from the field of topography: In a topographic A simple region merging method begins with the
surface, the watersheds are the lines dividing two assumption that each pixel is a separate image region or
catchment basins. A gray-scale gradient image is belongs to a pre-segmented region. Adjacent regions are
considered as a topographic surface where the value of then merged based on the region's statistics, such as the
each pixel represents the elevation or altitude at that point. mean [7]. Fast implementations of this model group pixels
Thus, high-gradient image edges represent watersheds by tracing "waterfalls" to their corresponding minima and
while low-gradient regions correspond to catchment then merge the resulting regions based on the depth or
basins. Watershed regions or catchment basins of the volume of the resulting pool [8]. Moreover, some "visually
'topographic' image are homogeneous meaning that the heterogeneous regions" may be "statistically homogeneous
corresponding pixels are connected with a path of regions". To overcome this problem, careful intervention
monotonically decreasing altitude. Each catchment basin is from the user or explicit prior knowledge on the image
represented by the set of pixels that will drain to the same structure is necessary for region merging.
local minimum. A watershed line will separate each pair of

Issam Dagher: Associate Professor, University Of balamand. Lebanon


Kamal El Tom: MS student. University Of balamand. Lebanon
Email: dagheri abalamand.edu.lb

1-4244-1 380-X/07/$25.00 ©2007 IEEE


2) Multi-scale Hierarchical Segmentation: Finding a curve for which E has a stationary value is a
One way to avoid the segments that result from small problem of variational calculus and one can apply the
variations in the gradient image is to low-pass filter (blur) fundamental Euler-Lagrange Differential Equation giving
the image in order to remove these gradients. However, the following equation:
low-pass filtering will also decrease the precision. In a(s)vss- (s)vssss- V E ext 0 (2)
multi-scale hierarchy [6], as the Gaussian scale sigma
increase, the watersheds change both number and position. Where the third term is normalized by the external energy
This means that objects constructed by merging high scale weight y. Solution of the above equation gives us our
segments, will have borders that are distorted compared to final contour minimizing E snake' One way to interpret this
the low scale borders, and this might cause problems when equation is to consider it to be a force balance equation of
shape information is used to control the merging of a system:
segments. Since segments are linked over scale this is
handled by replacing all high scale segments by the union F int + F ext 0 (3)
of its low scale "children" segments linked to it [6].
where
3) Marker-Controlled Watershed Segmentation: Fint = a(s)vss -(S)Vssss (4)
A marker is a connected component belonging to an
image. Internal markers associate with objects of interest, and
and external markers associate with the background. The F ext -VE ext (5)
criteria for an internal marker can for example be that it The final solution [9] is:
has to be surrounded by pixels with higher altitude. These
markers are then made to be the only allowed minima in
v t = [A +,r I] 1[,
v t_1 - F (v t_1 )] (6)
the watershed algorithm. The watershed lines created here Where t is the time step.
are defined as external markers. The external markers Splitting back v into its x and y components:
efficiently partition the image into regions containing one xt = [A +r I] '[ xt_l - Fx(xt-, )] (7)
single internal marker and some background. The problem
is then reduced to partitioning each of these regions into Yt =[A +rI] [ t Fy (xt- 1, yt- )] (8)
two, which can be done by some simpler segmentation F x (x t-I Y t-I ) is the gradient force in x direction on point
algorithm. ,

(x t-I tY-y ) of the image. It is retrieved from the gradient


I

B. Snakes image data when snake moves on to this point. Similarly,


In 1987, Kass, Witkin and Terzopoulus introduced the every point on the snake will follow this process and
Active Contour Models commonly known as Snakes [9]. moves to its new position. This is the complete minimizing
Active contour is defined as an energy minimizing spline. process for snake deformation.
Its energy depends on its shape and location within the
image. A snake can be considered as a number of control Some of the problems faced by the Snakes are:
points or snaxels that are linked together and free to
deform under the constraining forces. The control points 1) Initialization:
v(s) = [x(s), y(s)] are traditionally placed near the edges of A snake is traditionally initialized by hand (i.e. placing the
interest because of the poor capture range of a snake. initial positions of control points) and not automatically
Snake deformation is carried by minimization of an done by computer. The control points need to be first
energy function so that the contour will move from an positioned near the edges of interest due to the small
initial position until the energy will stabilize at significant capture range of image gradient. This is easily done by
edges. humans but is not an easy task for a system to intuitively
Further more, a snake works better when all the points realize the object of interest. However, if there is enough
are equally spaced and not too far apart. This is to ensure a computer power, one can always execute many snakes all
better performance in edge tracing. Therefore, an together and let it randomly pick up all the edges in an
interpolation mechanism is needed to set the appropriate image and then select the required ones.
limitations on the distances between control points. If
points are too close or too far apart, the mechanism must 2) Generating Snake Parameters:
delete or add points respectively in order to have a good Producing a suitable set of parameters for snake
initial position. At each step of deformation, this process of initialization is another difficulty. Parameters/weights of
checking the initial coordinates should be done to ensure energies totally control the performance of snake
better performance of a snake. deformation. Choosing a set of parameters (a, /3, y) that
The total snake energy is defined as [9]: can be used for a particular object in an image is a difficult
task for a computer. However, after initializing a snake, it
E snake 2 y[a(s)lvsl +(s)vs ] E ext (v(s)) ds is difficult to have a system that automatically changes the
parameters with respect to time using data from the image.
In Balloon model, the original snake internal force is
3) Capture Range: kept the same while a new external 'pressure' force term is
If an object in the image has an edge which shows a added to the snake force function.
concavity then the traditional snake will have a problem in The total balloon snake force is now:
tracing the edge at this concaved part. The reason is that F snake (F int + F ext ) + F pressure (9)
the capture range at these parts of edges is too far from the The external force is:
snake (Fig. 1). A poor solution is to increase the 'c ' value
of the Gaussian blurring filter, thus increasing the snake F ext =-VE ext
capture range, but the edge of the object will be also
blurred and cannot be traced accurately. A method called
= -G6 (X,y)® f(x, y)
'Gradient Vector Flow' (GVF) is lately proposed to FEk image (10)
overcome this problem. This method makes use of the VEimage
amplification of image gradient to increase the snake
capture range and deformation speed but should contain
the medial axis of the object [10]. Another method to solve
this problem is the Balloon Snake that will inflate or The pressure force added is given by:
deflate the snake from its farthest positions to the final F pressure = kln(s) (1 1)
edge locations thus increasing the snake capture range.
where n(s) is the normal unit vector to the curve at point
v(s), and k1 is the amplitude of the force, it indicates
inflation (when positive) or deflation (when negative).
At the same time, the value of k1 has an influence on the
Fig 1. Low capture range of traditional snake speed of deformation.
F snake (F int + F ext) + F pressure
Previous combinations of watershed and snakes have not
solved the over-segmentation problems. In [11], a snake
zone is defined around the object boundaries where the F VEima-
kim ag e
k 1n(sV) (12)
corresponding watershed points are candidates for the true
object boundary. Energy minimization or the searching The F snake is the force equation that needs to be
process is carried out using dynamic programming.
However, in most applications the difficulty is mainly in minimized in balloon snake deformation. Due to the newly
the definition (localization) of the snake zone. added pressure force, balloon model is able to overcome
The watershed-based segmentation on a down-sampled the over shrinking problem in the original snake. Also, if
image is used to get the initial contours for the snake [12]. the object has a troublesome shape, placing a balloon
This will guaranty automatic initialization of snakes using inside the image and expanding its contour will solve
watershed and smoothness of watershed lines using active matching the desired shape. The final solution is:
contours.
A new energy functional is derived based on the
vt = [A +± I] x [r v t_1-F(v t_1 )+F pressure] (13)
topographic distance definition of the watershed line [13].
This segmentation will incorporate control over the
smoothness of the watershed lines.
Smoothness of watershed lines is solved but the important II. WATERBALLOON SNAKE
over and under-segmentation problems of watershed are A. The Hybrid Technique
not addressed in [12], [13]. Since watershed regions are closed and disjoint regions
whose unions constitute the whole image and since
C. Balloon Snakes traditional snakes need to be initialized near the target
Laurent D. Cohen [14], introduced a model that makes the edges, the watershed lines around each region can be
curve behaves like a balloon with an additional external considered as initial snakes. Thus, watershed regions are
force which moves the contour out (inflation) or in involved in a multiple snakes approach. The idea of this
(deflation) along its normal thus eliminating the need for method is that the appropriate internal and external forces
the initial curve to be close to the correct edges. By placing should drive those snakes in order to be dragged by the
balloon snake inside/outside the object of interest and significant edges. Traditional snakes could not be
inflate/deflate contour to its boundary, this new model is initialized far away from the target edges. GVF snake
very useful for some object that is difficult to locate or too should contain the medial axis of the object therefore they
complicate to trace by original model. The Balloon snake cannot drive watershed lines.
passes over relatively weak edges and it stops on salient On the other hand, Balloon snakes can be inflated or
edges [14], [15]. deflated until they are attracted by the object edges.
Balloons are free to be initialized anywhere inside object
due to the powerful external pressure energy term that will
drive the snake to the correct location. Since watershed
regions are over-segmented version of the object, a
Balloon snake initialized at the location of a watershed
region contour will be a strong candidate for capturing the
salient object edges. Watershed regions cannot cross the
object edges, so Balloon snakes will not cause any
problem. They also don't have to include the object medial
axis, which ensure free initialization.
Figure 2 shows the block diagram of the proposed hybrid
segmentation technique. The original image is low-pass
filtered or blurred to remove small noisy regions. Then, the
watershed transform is applied on the gradient image to
segment watershed regions. Now, contour-tracing
techniques can be applied to extract watershed region
contours. Those contours will constitute the initial balloon
snakes that will deform to capture the object edges.
Balloon snakes have a large capture range so correct snake
deformation can be guarantied even if the initial snake (or (c)
the contour of the watershed region) is far away from the Fig. 3. Proposed Segmentation Strategy. (a) Initial watershed regions
target edges. This situation is frequent for a given and lines (over-segmented image). (b) Deformation of initial watershed
lines with Balloon Snakes. (c) Final Contours with the two blood cells
candidate watershed region, so balloon snake is most separated (lines are drawn inside pixels).
suitable to drive the watershed contours. Once a snake
captures one object, the image is relabeled so that the next
snake will not be a watershed contour inside the captured B. Optimization of snake parameters based on Energy
object. For each snake, parameter optimization is required Tracking
to ensure optimum edge localization.
This process is repeated for each snake to obtain the A snake may not be located at the target edges due to its
WaterBalloon segmentation. Illustration of the technique sensitivity to the snake parameters. Thus there is a need to
on the two touching blood cells is shown in Figure 3 with a parameter optimization that best localize the correct edges.
low-resolution image. The lack of knowledge about those edges is what makes
optimization of the snake parameters a very hard problem.
Thus a new approach for determining those parameters is
proposed that is derived from the energy functional
behavior of WaterBalloons.

Fig. 2. Block Diagram of WaterBalloon Segmentation.


-800 3000

-iooo -,, 2500

-1200 2000

-1400 F 1500

-1600
1000
-1800
500
-2000

-2200
-500 .\ \
-2400
-1000 t,t
-2600
-1500
-2800 L 0 5 10 15 20 25 30 35 40 45
0 50 100 150 200 250
Fig. 5. Total Energy term behavior when Balloon inflates outside object.
(a)
-200 :1
Figure 5 corresponds to the case where the balloon is
inflated outside the target boundaries. In this case, the
F400 4 l o W n ( r Total energy increases considerably. Whenever this case is
600 l E t b a o encountered, the parameters need to be changed and the
-800A
pressure weight should be decreased. This parameters
updating is continued until the energy transitions follow
-1000V '-,. X
the case of Figure 4. The optimum values obtained are then
-1200 -h fed back to deform the snake.
-1400
The block diagram of the optimization stage is shown in
50 0 100 150 200 250
Figure 6 where the parameters are initialized equally to
(b) give equal importance to all energy terms.
Fig. 4. Energy plots of WaterBalloon Snakes. (a) External Energy term From the above discussions, a criterion can be tested to
behavior. (b) Total Energy term behavior.
check if there is an energy pulse with transition to a high
The Total Energy term is shown in Figure 4(b), the energy energy state and a back transition to a lower energy state or
behavior seems not to provide any information about the there is only a transition to a high energy state in case of a
snake performance. The challenge was to find a continuously inflating balloon. This criterion is based on
relationship between this energy behavior and the optimum the initial, final, and maximum values of the energy.
parameters.
if
Efinal
f
-Einitial
i T< , then there is an energy pulse
<
The key solution was to look at the total energy term and
the corresponding elasticity, bending, external and Emax
pressure (Balloon) energy terms. The idea is clear in and the balloon snake will
Figure 4(a) and (b) where two conclusions are drawn: stabilize (T = 0.5 is satisfactory
* There exist an energy transition to a high Energy for this application).
state and then a drop to a lower energy state. else
* The drop to the lower energy state occurs there is only transition to high energies with
simultaneously in both the external and total inflating balloon snake.
energy terms. Enlergy Trackinlg:
This is not by chance; as the edge will deform from its Initialize nryOptimize d
initial position at the weak edge to the strongest edge, it P aramet ers -hlimi rlrg
P aramet
iiie
ers

traverses a homogeneous region where gradient field is Equally K Enler gy S;tate s


weak and no edge is encountered.
In this homogeneous region, the external energy term is
maximal (in absolute sense) and thus the total energy term | Parameters
is maximal. This explains the transition to a high-energy Fig. 6. Block Diagram of Optimization with Energy transitions tracking.
state in both the total and external energy curves.
One can also note that the final energy state is lower than C. Application to blood Cell Image
the initial one since the snake is minimizing its total
energy. Figure 7(a) shows red blood cells image with two touching
This behavior will be the typical for the proposed pair of cells. In traditional segmentation methods, touching
WarterBalloon technique: starting from a watershed line as objects (the blood cells) are considered as one object
an initial balloon snake, the deformation from the weak because these methods are insensitive to the low intensity
edge to the strong edge through the homogeneous region levels of the border pixels between these objects.
will always induce this energy pulse. In Fig. 7(b), watershed transform was applied to the
gradient image. In a sequential manner, a watershed region
is picked up and will be subject to balloon snake
deformation until an object is captured. Next, the image is
relabeled to ensure that the second watershed region will
not fall in the already captured object. Once a watershed
region deformed with the balloon snake captures a single energy transitions tracking, and optimum edge
cell part of two or more touching cells, the following localization is achieved.
deformations will separate the remaining cells (Fig 7(c,
d)). Water-Balloon Snake was able to separate the REFERENCES
touching objects. [1] Gonzales R. C. & Woods R. E. Digital Image Processing, ISBN: 0-
201-18075, 3rd edition, Prentice Hall 2002.
[2] Serra J. Image Analysis and Mathematical Morphology. Academic
Press, London 1982.
[3] Vincent L. & Soille P. Watershed in digital spaces: an efficient
algorithm based on immersion simulations, IEEE Trans. on PAMI,
vol. 13, no. 6, pp.583-598.1991
[4] Beucher S. & Meyer F. The morphological approach to
segmentation: the watershed transformation. In E. R. Dougherty,
editor, Mathematical Morphology in Image Processing, chap. 12,
pp. 433-48 1. Marcel Dekker, New York. 1993
[5] Najman L. & Schmitt M. Geodesic saliency of watershed contours
(a) and hierarchical segmentation, IEEE Trans. on PAMI, vol. 18, no.
12, pp. 1163-1173 Dec. 1996.
[6] Gauch J. M. Image segmentation and analysis via multiscale
gradient watershed hierarchies, IEEE Trans. on Image Proc., vol. 8,
pp. 69-79. Jan. 1999
[7] Bleau A. & Leon L. J. Watershed-based segmentation and region
merging, Computer Vision and Image Understanding, vol. 77, pp.
317-370 2000.
[8] Haris K., Efstratiadis S.N., Maglaveras N. & Katsaggelos A.K.
Hybrid Image Segmentation Using Watersheds and Fast Region
Merging. IEEE. Trans. on Image Processing, vol. 7, no 12, pp.
1684-1699 Dec. 1998.
[9] Kass M., Witkin A. & Terzopoulos D. Snakes: Active contour
(b) models. International Journal of Computer Vision, vol. 1, pp. 321-
331 1987.
[10] Nilanjan R., Acton S.T., Altes T., de Lange E. & Brookeman J.
Merging Parametric Active Contours within Homogeneous Image
Regions for MRI-Based Lung Segmentation. IEEE Trans. on
Medical Imaging, vol.22, no. 2. Feb. 2003.
[11] Park J. & Keller J.M. Snakes on the Watershed. IEEE Trans. on
PAMI, vol. 23,
no. 10 Oct. 2001.
[12] Kiran V. & Bora P.K. Watersnake: Integrating the Watershed and
the Active Contour Algorithms. IEEE. 2003.
[13] Nguyen H.T., Worring M. & van den Boomgaard R. Watersnakes:
(c) Energy-Driven Watershed Segmentation. IEEE Trans. on PAMI,
vol. 25, no. 3. March 2003.
[14] Cohen L.D. On Active Contour Models and Balloons. Computer
Vision Graphics and Image Processing: Image Understanding, vol.
53, no. 2, pp. 211-218 March 1991.
[15] Cohen L.D. & Cohen I. Finite-Element Methods for Active Contour
Models and Balloons for 2-D and 3-D Images. IEEE Trans. on
PAMI, vol. 15,no. ll,pp. 1131-1147Nov. 1993

JU)
Fig. 7. Water-Balloon Segmentation. (a) Original Image. (b) Watershed
Regions. (c) Balloon Snake deformation. (d) Water-Balloon Segmentation
with the touching cells separated.

III. CONCLUSION
In this paper, a hybrid Watershed Balloon snake
technique for image segmentation is introduced. The
new technique combines the use of Watershed
transform and Balloon snakes in an innovative
approach that detects salient objects in an image.
Waterballoons provide the advantage of reducing
over-segmentation of Watershed while preventing
under-segmentation problems. Touching objects are
separated, snake parameters are optimized based on

Anda mungkin juga menyukai